[Libreoffice-bugs] [Bug 46845] Iterations not converging in Calc

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=46845

--- Comment #8 from b. <newbie-02 at gmx.de> ---

for this and the plenty other questions / bugs regarding circular references /
iterative calculations: 

explanation of the calculation order used in ex$el: 

https://stackoverflow.com/questions/18084718/calculation-order-when-iterative-calculation-circular-reference-is-enabled

checked with ex$el 2010, works that way, 

is there a similar description for calc? 

differences: 

for a recalc - F9 - excel does a round as described above, while calc looks
like start at the cell in focus, and stops after one round? (after which
definition? re-raeching that cell?) thus some calculations give different
results,  

ex$el does show 'invalid results' (results not converging and being off limt of
'minimum change') after the set number of steps, while calc shows err:523 if
not reached the change limit, 

(try second variant of first example from
https://www.spreadsheetweb.com/excel-iterative-calculation/

there are also samples to use iterations for timestamping ...) 

from C3: '=C4' and C4: '=C3+1' resulting in identical values (after hard recalc
with one iteration and minimum change > 1) and results being different by one
if you exchange the cells, i'd conclude that calc calculates from bottom to
top, 

from C3 '=D3' and D3: '=C3+1' resulting in identical values (after hard recalc
with one iteration and minimum change > 1) and the left cell being one more
than the other if you exchange them, i'd conclude that calc calculates from
right to left, 

from C3 '=D2' and D2: '=C3+1' resulting in identical values (after hard recalc
with one iteration and minimum change > 1) and D2 being one more than C3 if you
move it to B4, i'd conclude that calc first calculates the rightmost column
from bottom to top, and then advances left to the next columns, 

so there are simple but significant differences between the schemes of Ex$el
and calc to solve circular references, since the choice of the starting point
and the calculation path are important for the output of iterative calculations
it is normal that Ex$el and calc give different results, 

big question: 'do we find our own way right (and if so why 'more right' than
Ex$el), or do we want to work more towards Ex$el compatibility? 

and! 

who dares to tackle this 'hairy soup'? (cited from erAck)

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20200802/40c7add0/attachment.htm>